Robust response abacus

[aarondr77]

Description

• Fixes Abacus response format after this conversation with Abacus CTO
• Updates our prompts to use valid json objects in all of the examples

Testing

Please provide a list of the ways you can "access" or use the functionality. Please try and be exhaustive here, and make sure that you test everything you list.

• I have tested this on real data that is reasonable and large
• If I changed the interaction with JupyterLab, I tested that it does not break other programs (like VS Code), and tested that it works "multiple times" in the same notebook.

Documentation

Note if any new documentation needs to addressed or reviewed.

---

Note

Medium Risk
Touches LLM response formatting and tool-invocation prompt contracts; mistakes could cause malformed responses or tool parsing failures across providers, though changes are small and localized.

Overview
Fixes Abacus JSON response formatting by switching the Abacus response_format from {"type":"json"} to {"type":"json_object"} in open_ai_utils.get_open_ai_completion_function_params, while keeping full JSON-schema formatting for non-Abacus providers.

Hardens agent prompt examples to valid JSON by updating tool/example payloads (e.g., cell_update, run_all_cells, scratchpad, ask_user_question, finished_task) to use double-quoted keys/strings and consistent schemas (including adding fields like code_summary/cell_type in the smart debug example).

^{Written by Cursor Bugbot for commit 9189e7f. This will update automatically on new commits. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
monorepo	Ready	Preview, Comment	Feb 18, 2026 9:38pm